Comparison of Subspace Projection Method with Traditional Clustering Algorithms for Clustering Electricity Consumption Data

There are many studies about using traditional clustering algorithms like K-means, SOM and Two-Step algorithms to cluster electricity consumption data for definition of representative consumption patterns or for further classification and prediction work. However, these approaches are lack of scalability with high dimensions. Nevertheless, they are widely used, because algorithms for clustering high dimensional data sets are difficult to implement and it is hard to find open sources. In this paper, we adopt several subspace and projected clustering algorithms (subspace projection method) and apply them to the electricity consumption data. Our goal is to find the strength and weakness of these approaches by comparing the clustering results. We have found that traditional clustering algorithms are better to be used for load profiling by considering global properties and subspace or projected methods are better to be used for defining load shape factors by analyzing local properties without prior knowledge. Keywords-subspace projection; traditional clustering; Kmenas; SOMs; Two-Step; local property; global propert.